Abstract
The specificity of neural responses to visual objects is a major topic in visual neuroscience. In humans, functional magnetic resonance imaging (fMRI) studies have identified several regions of the occipital and temporal lobe that appear specific to faces, letter strings, scenes, or tools. Direct electrophysiological recordings in the visual cortical areas of epileptic patients have largely confirmed this modular organization, using either single-neuron peri-stimulus time-histogram or intracerebral event-related potentials (iERP). In parallel, a new research stream has emerged using high-frequency gamma-band activity (50-150 Hz) (GBR) and low-frequency alpha/beta activity (8-24 Hz) (ABR) to map functional networks in humans. An obvious question is now whether the functional organization of the visual cortex revealed by fMRI, ERP, GBR, and ABR coincide. We used direct intracerebral recordings in 18 epileptic patients to directly compare GBR, ABR, and ERP elicited by the presentation of seven major visual object categories (faces, scenes, houses, consonants, pseudowords, tools, and animals), in relation to previous fMRI studies. Remarkably both GBR and iERP showed strong category-specificity that was in many cases sufficient to infer stimulus object category from the neural response at single-trial level. However, we also found a strong discrepancy between the selectivity of GBR, ABR, and ERP with less than 10% of spatial overlap between sites eliciting the same category-specificity. Overall, we found that selective neural responses to visual objects were broadly distributed in the brain with a prominent spatial cluster located in the posterior temporal cortex. Moreover, the different neural markers (GBR, ABR, and iERP) that elicit selectivity toward specific visual object categories present little spatial overlap suggesting that the information content of each marker can uniquely characterize high-level visual information in the brain.
Introduction
Visual neuroscience has long established the brain's modular nature. Functional magnetic resonance imaging (fMRI) has identified specialized regions like the Fusiform Face Area (FFA) that respond selectively to specific categories. However, fMRI provides only an indirect measure of neural activity via blood flow. To decode the brain's actual electrical language, researchers examine various electrophysiological markers: Intracerebral Event-Related Potentials (iERPs), high-frequency Gamma-Band Responses (GBR, 50–150 Hz), and low-frequency Alpha/Beta Activity (ABR, 8–24 Hz). A critical unresolved question is whether these different signals—voltage blips, local high-frequency oscillations, and lower-frequency rhythms—coincide spatially or represent distinct processing channels.
Methods
This study utilized intracranial EEG (iEEG) recordings from 18 epileptic patients with 1,760 implanted electrodes, offering millisecond-level temporal resolution across the temporal, frontal, and occipital cortices. Patients performed a visual task viewing images from seven categories: faces, scenes, houses, consonants, pseudowords, tools, and animals.
Researchers applied Hilbert transforms and Wavelet decompositions to extract frequency-specific amplitudes. A Selectivity Index (SI) quantified electrode "pickiness," while signal detection theory (ROC/AUC) assessed the ability to decode stimulus categories from single-trial neural data.
Results
The Spatial Dissociation of Neural Markers
The most significant finding was a striking lack of spatial overlap between different neural markers. Despite both GBR and iERP showing high category-selectivity, they rarely occurred at the same recording sites. Brain-wide, the spatial overlap for category-selectivity between GBR and iERP was only 3.8% (and only 2.9% within the temporal cluster). This suggests that the brain utilizes distinct physical locations and "channels" (frequencies vs. potentials) to encode the same high-level visual information.
Decoding and Dynamics
Both GBR and iERP proved to be high-fidelity carriers of information, with AUC values frequently exceeding 0.8, allowing for accurate single-trial stimulus decoding. While both signals peaked rapidly (100–200 ms post-stimulus), GBR tended to be "stickier"—remaining elevated longer than the transient iERP blips, potentially reflecting sustained processing or representation maintenance.
Distribution of Selectivity
While selectivity was most prominent in the posterior temporal cortex (aligning with fMRI modules), category-specific responses were distributed broadly, including within the frontal lobes. Of the responsive sites, 16.9% responded exclusively via GBR, 2.7% via iERP, and 14.5% via ABR.
Discussion
The observed dissociation between GBR, ABR, and iERP suggests that each marker uniquely characterizes high-level visual information. This may be due to different scales of neuronal synchrony, where GBR reflects local firing and iERP/ABR reflect broader network dynamics.
Limitations include the sparse sampling inherent to clinical electrode placement and the use of a clinical population. Future work should investigate the functional "cross-talk" between these dissociated signals to determine if activity in one marker triggers or modulates another across different brain regions.
Summary updated